A Fully Integrated Model for the Optimal Operation of HydroPower Generation by Francois Welt University of Toronto, Dec. 4, 2012 01/2012 2 Hatch Power and Water Optimization Group • Engineering Company • Specialized group within Hatch Renewable Power • Experience: – Over 40 systems implemented – Experience with different types of hydro systems • Supported by over 9,000 multi-disciplinary engineering professionals worldwide 01/2012 2 3 Hatch Power and Water Optimization • Water Resource and Power System Modeling – Simulation and Optimization • System Implementation – Configuration, Test – Integration / Communications – Install and Train • Studies • Asset Management / Life cycle analysis • Wind Farm Design Optimization 01/2012 3 4 Columbia Vista - Integrated Optimization Model 01/2012 5 Hydro Optimization in Generation Planning Concepts • Make best use of limited hydro resources • Meet operational constraints • Maximize Profits – Maximize sales/ Minimize costs – Calculate optimal plant/unit MW – Calculate optimal WL trajectory/ spill releases – Calculate bid curve Optimization technologies becoming increasingly attractive with improvements in computing speeds/ capabilities 01/2012 5 6 Optimization Statistics Examples of potential economic benefits from optimization - Short term operation Ref: “Assessing the Economic Benefits of Implementing Hydro Optimization”, Hydro Review magazine, 1998 1.4 1.2 1 0.8 0.6 Typically, potential improvements between 1 – 5% 0.4 0.2 0 Market Spill Efficiency Head 01/2012 6 7 Hydro Optimization Time Scale Plant/ units Real Time (RT): Dispatch Hour/day end/ Smaller reservoir Short Term (ST): • Schedule • Transactions To end of week/month Larger reservoir Long Term (LT): • Generation/Water Plan • Targets and Water Values To end of water year 01/2012 8 Optimization Problem Must formulate problem in terms of: •Objective functions •Constraints •Rules of operation •Physical relations •Decision Variables Objective Max{ [Re venues( X ) Costs( X )]} Time Constra int_ F ( X ) 0 Characteristics: •One set of decisions per time step, piece of equipment •Hydraulic network •Transmission network •Large problem size 01/2012 8 9 Physical Representation • Hydraulic Network – – – – – – Source: Inflow Points Sink: downstream outlet Water conveyance/ Flow Storage Head (Potential energy) and head loss Can be bi-directional (gen/pump) • Electric Network – – – – Source: Generation points Sink: Load or Market points Bi-directional Energy losses 01/2012 9 Hydro System Components River Reach 10 Inflow Arc Reservoir Node Power Arc Spill Arc Tailwater Junction Node River Reach 01/2012 11 Columbia River System 01/2012 12 SCE Vista Big Creek Hydro System Representation 01/2012 13 Generation Resource (Hydraulic flow to Electric MW) 01/2012 14 Controlled and Uncontrolled Spillways 01/2012 15 Rock Island Schematic Powerhouse Two Powerhouse One 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 Spillway F F F 01/2012 16 Hydraulic Network Representation – Continuity equation at each reservoir node Σ Qin – ΣQout = V(t) – V(t-1) – Continuity equation at each junction node Σ Qin – ΣQout = 0 – Conveyance in reach arc Qout (t) = Σ α(n).Qin(t-n) 01/2012 1 17 Transmission Area Load Demand Committed Transactions X Market • Purchase • Sale Bilateral • Purchase • Sale D Transaction X H Hydro Generation T Thermal Generation Wind Wind Generation 01/2012 18 “aggregate unit” Bus Configuration COB2P Contract Bus BPA1 X(COB2)P Supply Area River 1 Plant 1 Bus A COB1 X MW Y MW X(COB1)S X(COB1)P X(COB2)S COB2S Bus B W MW X(PNW1) S X(PSE) S X(PNW1) P X(PSE) P COB3S X(COB3)S L X(PNW2) S X(PNW2) P “line limits” X(BPA-X) S BPA3 X(PNW3) S X(PNW3) P Z MW X(MIDC) S River 2 Plant 4 X(MIDC)P MID-C “group line limits” Plant 2 Plant 3 BPA2 COB3P X(COB3)P PSE River 3 Plant 5 Plant 6 01/2012 19 LT Vista Physical Model Transmission System 01/2012 20 Network Representation • Electric Network – Continuity equation at node (bus) Σ MWin – ΣMWout = 0 – Losses through conveyance (tieline) MWout = Mwin - α.Mwin^2 • MW Energy • MW ancillary service (reserve) 01/2012 2 21 Physical Representation Reserves and Generation • Unit/Plant Balance Equation NON-SPINNING OPERATING RESERVE NON_AGC TOTAL SPINNING LOAD FOLLOW. CONTROL SPINNING AGC (Regulating) REG Down MW GENERATION Plant MAXMW Plant MW Re serves 01/2012 23 Joint Optimization – Energy and A/S markets with price forecasts – Optimal trade-off between energy and A/S • Spin • Non Spin • Regulation Up • Regulation Down Energy •Unused capacity can earn revenues with resulting unused water still sold as energy at a later date •Some of the unused capacity can be converted into energy when reserve is called (Take) 01/2012 24 Physical Relations: Plants and Units • Power-flow-head relationship (3-D) 01/2012 25 Spillway Equations Q = Co · Le · Open · (hwl - E)Eo E = Esill or twl hwl Free Overflow hwl ESill Submerged Flow twl ESill Q = Cf · Le · (hwl - Esill)Ef 01/2012 26 Operational Constraints Representation • Hydraulic Constraints – Simple Constraints on Flow, storage (WL), MW – Time aggregated constraints (linear) • Max average • Max/min between periods – Relational constraints (including step functions) • Electric Constraints – Simple Max/ Min on generation – Tieline flow (congestion) – Reserve (min/max) 01/2012 2 27 Operational Constraints Representation 01/2012 28 Complexities in Formulation • Uncertainty – Inflow – Load – Market price Long Term • Hydraulics – Non-linear physical constraints • generation with cross product (flow * head^a) • Losses (quadratic) • Spill representation – Spatial/time connectivity Short Term • Discreteness – Start/stop costs – Spinning reserve – Non continuous operating range Real Time • Large Scale – Time dependent decisions (up to 200,000 decision variables / constraints) 01/2012 29 Preferred Schemes for Hydro Linear Programming •Piecewise linearization •Successive Linearization •Semi-heuristics • • Decomposition •Subproblems •Bender’s cuts •Dynamic Programming •Nonlinear Programming Plants are hydraulically and electrically connected – Water conveyance – Load, reserve Fixed amount of water over time – strong temporal interdependency 01/2012 2 30 Long-term Planning 01/2012 Long Term Model Principles 31 • Consideration for Future Uncertainty – Stochastic • Detailed Physical Representation • Simplified Time Definition – Periods (week(s), month) – Sub-period (peak, off-peak, weekend,…) • Time Average answers • Based on scenario analysis – consider all cross correlations 01/2012 32 LT Vista Mathematical Model Hydraulic network: – arcs (plant, spill, river reaches) – nodes (reservoirs, junctions) Engine Stochastic SLP (2 stage) Detailed Plant Operation Detailed constraint set Benders Decomposition • Electric network: - Buses tielines Inputs: - hydraulic: stochastic inflow, start/end WL - electric: transactions, load Constraints: - hydraulic (flow, elevation, etc.) - electric (transmission, etc.) 01/2012 LT Vista Methodology 33 • Two Stage LP • Decomposition Master 1st Period / Future period subproblem Future 1 Future 2 NOW Future 3 Future N 01/2012 LT Vista Methodology 34 • Multi-dimensional Uncertainty -- Inflow, Market and Load Market Hydrology H1 H1_M1 H1_M2 H1_M3 H1_Mm Load H1_M2_L1_ H1_M2_L2 H1_M2_L3 H1_M2_Ll 01/2012 35 LT Vista Time Definition • Period: – basic model time step (e.g., 1 week) • SubPeriod: – Peak-off peak (Load duration) aggregation within periods • Time blocks – constraints tying several periods/subperiods subperiods period Time block 01/2012 36 LT Vista Display – Probabilistic WL 01/2012 LT Vista Display – Probabilistic MW 01/2012 37 38 Short-term Scheduling 01/2012 Short Term Model Principles • • • • 39 Deterministic Model Detailed Physical Representation Detailed Hourly Time Definition SLP numerical scheme with piecewise representation: – MW/Flow relation – Tieline losses • Unit Dispatch/Unit Commitment Subproblem – Nonlinear Programming – DP • Spinning reserve allocation subproblem • Integrated handshake with Long Term Model • Market Analysis 01/2012 40 Unit Dispatch/ Unit Commitment Subproblem • • • • Plant Representation based on optimal unit dispatch/ unit commitment around base solution Plant Generation function used in SLP Best Dispatch answers used in scheduling General LP problem formulation cannot deal with discrete decisions – unit ON or unit OFF Plant 1 Non continuous operation Plant 2 Unit Dispatch Model •Snapshot Non linear analysis •Fixed Head Plant N 01/2012 41 Spill Allocation • • • • • Aggregated spill representation Piecewise linear representation No flow zone Sequencing issues – heuristic vs integer set Stability issues Spill 1 Spill 2 Spill N 01/2012 42 LT – ST Handshake • Type – Economic • Seasonal Reservoirs: Value of water in storage applied to end of opt period water levels • Other Reservoirs/head ponds: Max Target Water Levels at the end of opt period. – Target Water Levels • Seasonal Reservoirs: LT Target Levels applied to end of opt period water levels • Other Reservoirs/head ponds: Max Target Water Levels at the end of opt period – Target Flow Releases • Seasonal Reservoirs: LT Target Levels applied to end of opt period water levels • Other Reservoirs/head ponds: Max Target Water Levels at the end of opt period • Others – meet target water levels defined by user • Custom – Combination of above 01/2012 43 Spinning Reserve Allocation Subproblem • Linearized formulation of spinning reserve • Subproblem is to find best unit allocation to meet spinning reserve requirements • LP Unit representation 120 Reserve 100 80 Operating Spin Spin + Reg Down 60 40 20 0 0 20 40 60 80 100 120 MW Gen 01/2012 44 Reduction of Problem Size: User Defined Time Grouping 01/2012 45 Total Bus Generation: Comparison between Time Groupings 4 hr 2 hr 8 hr 01/2012 Reduction of Problem Size: ON/OFF River Status 46 01/2012 47 ST Vista Run Times (866 MHz) Total Study Time 35.0 Time (minutes) 30.0 Cold Starts 25.0 Day Ahead Study Period 20.0 15.0 10.0 Hot Starts 5.0 0.0 0 20000 40000 60000 80000 # of Constraints (row size) 01/2012 48 Semi-Heuristic Resolution Schemes • • • • • Plant retirement/commitment Plant zone resolution Uncontrolled spillway structure Semi-heuristic – does not cover all solution space Perturbation to the LP global problem M W Flow 01/2012 49 Price-Volume Curves Methodology • Cost sensitivity calculation Future MW Base Storage Dev $ Time MWh 01/2012 50 Real Time Dispatch 01/2012 Real Time Model Principles 51 • • • • Deterministic Model Detailed Physical Representation Detailed sub hourly Time Definition Detailed Unit Dispatch/Unit Commitment Subproblem • Integrated handshake with Short Term Model 01/2012 52 Unit Commitment – Dispatch Rules • • • • • Minimum unit run time Minimum unit down time Maximum number of unit state changes in one time step Unit start / stop costs Dynamic unit status eligibility • • • • Unit availability Unit available for start Unit available for shutdown Unit fixed operations • Chosen algorithm – Dynamic Programming optimization 01/2012 53 Unit Commitment – DP Formulation 01/2012 54 Unit Commitment – DP Features • Only states derived from every time step, snap-shot, unit dispatch results are considered • Only eligible state paths are considered • Two cost components are evaluated • State transition costs ( unit start / stop costs ) • State operation costs ( cost of water to meet generation requirements ) • Objective function – minimize total dispatch cost 01/2012 55 Efficiency Gains After Before 01/2012 Conclusions 56 • Future Trends and Developments – Quality of Short Term Schedule • Robustness/stability – Expansion of market analysis – Handling of uncertainty in Short Term scheduling – Higher flexibility/performance in LT stochastic analysis 01/2012